Recent developments in telecommunications technologies have allowed expansion of service offerings from the ubiquitous voice telephone service model to include an array of packet data communication services. Packet data service offerings are rapidly migrating from narrowband technologies to broadband technologies, to facilitate various multimedia applications. This evolution of data communication technologies has included a rapid development and deployment of wireless mobile communication networks. Carriers are now deploying wireless network technologies offering broadband packet data communication services. Packet data communication networks utilize various techniques to control access to or to allocate scarce network resources, such as packet communication channels in a wireless network.
Transmission deference is the notion of withholding the transmission of a packet over a network link until transmission conditions are more favorable for reliable data transfer. Since often a source output data rate cannot be controlled by the system, the generated data needs to be transmitted to its peer entity within a certain predetermined time window interval. Data being received outside this window often is discarded as useless. Other times, data delay constraints are rather relaxed. Data again has to be transmitted within some time period, but that period could be much longer than that of low delay constrained data. Part of the art in communicating is to relay all sorts of different delay constrained data concurrently at maximum throughput without violating the data delay constraints. Obtaining a high data throughput when operating with highly uncontrollable sources requires buffering and sophisticated transmission scheduling algorithms. Link or network resource allocation algorithms often perform highly complex optimization procedures in determining how to allocate link or network resources to different users. For long-term resource utilization, enough time must be available for collecting resource demands, processing them and then performing the resource allocation.
In highly centrally controlled networks, a centrally located controller controls resource allocation; and all data transmissions are scheduled in advance. When enough time is available for carrying out the underlying processes, these types of resource allocation mechanisms are highly effective and tend to maximize system throughput while providing the quality of service expected by the system users. Systems of this type are mostly seen when the performed services are low delay constrained, such as in voice or real time video applications. Here, strict delay constraints are imposed and enough system capacity is allocated to guarantee the required QoS. Since the time at which the voice or video connection is actually implemented in the network is not highly constrained, centrally located controllers have enough time to receive a request, process it, allocate the required network resource and inform the user that the resources have been allocated for use. Some form of central controllability provides a sense of system reliability; and unexpected or even catastrophic events can be dealt with efficiently and effectively within a short time interval. Service providers often negotiate contracts with their customers, which guarantee that events of this type either do not happen, or their effects are kept to a minimum.
Until now, real time services have been offered over networks, which would guarantee their required QoS. For other types of services, data networks operate in parallel to offer services not requiring low delay constraints. The best example of such a network is the Internet. Building and monitoring these different networks at all times is a huge cost, which translates to higher usage fees, albeit at the required QoS. The experience of the Internet so far is that services are not guaranteed. The resource allocation in these types of networks are more or less non-existent, and when the system is loaded, the delay could become very long thereby rendering certain applications unusable.
Other types of networks that have lately seen widespread use are local area networks (LANs) and in particular wireless local area networks (WLANs). Here the notion of resource allocation is associated with the end user device rather than a centrally located controller, although certain types of such networks do retain a level of central controllability. With the end user devices becoming more intelligent, resource management has become a more distributed oriented process. Clearly, allowing a fully distributed resource allocation does have considerable merits. For instance, a centrally located and expensive node to perform resource allocation is not required. Furthermore, the delay associated with data collection, processing and other resource management signaling functions is removed.
The differentiation of services does not stop at real time vs. non real time services. With the proliferation of electronic devices, a variety of different services requiring different QoS levels have been introduced. It is certainly not economically sensible to build a different network for each set of relatively dissimilar services. A merging of different types of services onto a single network platform is required.
Having a single network for all services imposes a major hurdle in the resource allocation requirements. A centrally controlled system cannot respond fast enough to accommodate changes in the network. Usage and scheduling a large number of users presenting bursty types of traffic greatly increases the amount of signaling and control overhead, sometimes to the point that the amount of overhead is more than the amount of the transmitted data itself. Furthermore, the signaling could be long enough to eliminate one of the biggest advantages of centrally controlled systems, which is the low delay constraint of guarantee.
A distributed control communication system can respond very fast in providing resources to different end points in a highly dynamic manner. End points are allowed to make their own measurements and decide about allocating (Capturing) a network resource for themselves. Clearly, the delay for resource allocation could be made very small; and there could be very little if any associated signaling overhead. Clearly, if a resource is not in use, an end point entity, which needs it, should not have to ask the network and wait for possibly a long time period to get a response back when the resource has been free all along.
Methods have been developed which tend to perform well over a large range of network types or mixture of service usage. Methods such as pure Aloha and slotted-Aloha are the pre-cursors of the large variety of methods that have been developed over the years. In pure-Aloha, an end point transmits whenever it has data. If the transmission collides with a transmission from another end point, the data is transmitted again until a successful transmission has occurred. There is minimal other signaling overhead required (besides the acknowledgment of successful transmission). The throughput of this method, however, is below 18%. This is clearly not a good system when the medium the system is communicating on is expensive. A slotted Aloha system maintains a time slot discipline, where a user transmits only at the beginning of a time slot and may hold the slot for its duration and beyond. The throughput of this approach is twice as much as for pure Aloha, but this gain is often not substantial enough to be attractive. Both Aloha types offer no sensing of the medium before transmission.
Methods like carrier sense multiple access (CSMA) require the end points to perform a carrier sense by which they can determine if the medium is already in use, before they themselves try to use it. CSMA being simple and effective has been popularized in various existing WLAN systems. It can operate without a central control node, and it only requires two end points to set up a functional network.
Fully distributed controlled systems, however, have the drawback that vital system resource and usage data are not available to all end points at all times. Each end point needs to operate based solely on the amount of information it can retrieve from sensing its environment. Having to operate with less and often minimal information, often places distributed control systems at a great disadvantage. Another major impairment is that of observability. In broadly physically distributed networks, end points cannot observe all medium activity over the full network. This limited observability could impose severe operational penalties to the overall system capacity. Methods such as RTS/CTS have been developed to deal with this problem often termed the “Hidden Node” problem.
Often systems need to retain a central controller. Since most small networks need to interact with other larger networks, through some gateway, this gateway is a natural location for a central network controller. Having a central controller could offer various advantages to a system that otherwise uses mostly distributed control to perform resource allocation. When a flavor of central control is imposed on a distributed functionality, capabilities otherwise difficult to obtain now become readily available. A semi-autonomous resource allocation could provide the fast speed of distributed (autonomous) systems and the reliability and information access of central control systems. Resource allocation methods of this type are founded for example in ISMA (Inhibit Sense Multiple Access) systems. Here, information is relayed from the network about the current use of the network resources. An end point is inhibited from accessing a resource that the network has declared as unavailable. The end points could contend for the idle or available resources in a distributed manner.
The issues behind Distributed vs. Centralized Resource Allocation in mobile networks could be itemized as follows: In Distributed networks, a) wireless end points control their transmission themselves, b) resource allocation processes (algorithms) are simple, c) transmissions are robust against other radio interference, d) ability to operate in ad-hoc networks or when channel is shared. In Centralized Network, a) a central controller as a Base Station is required, b) a Base Station schedules both Uplink and Downlink transmissions, c) better control over radio resources, d) better service guarantees such as (fairness, delay, loss, etc.) e) better suited to commercial systems such as cellular and GPRS.